This paper describes the functionality of a local planner module for autonomous mobile robots. As an intermediate layer of a hierarchical planning system for autonomous vehicles, this module translates goals provided by a map-based planner into general vehicle action modes called activities. Activities are composed of pre-defined sets of reflexive behaviors which yield known performance characteristics under certain specific environmental conditions. Activity selection is performed on the basis of perception-supplied descriptions of the local environment, descriptions of expected terrain characteristics from a map-based planner, local map information accumulated by the local planner itself, and knowledge stored with each activity describing how that activity may achieve various goals and handle various failures. Local planner reasoning occurs within a monitoring process associated with the selected activity. Selected activities communicate with the local planner by posting messages onto a blackboard. These messages convey information to the local planner such as the distance to the nearest obstacle in a certain direction and activity completion status. As the vehicle moves, the local planner builds a map that records the path traversed, as well as major features and landmarks encountered. This information is used when the vehicle needs to backtrack. The local planner is implemented in Lisp, and has been demonstrated in a simulated environment.
Vincent S. Wong,
David W. Payton,
"Goal-Oriented Obstacle Avoidance Through Behavior Selection", Proc. SPIE 0852, Mobile Robots II, (1 January 1987); doi: 10.1117/12.968230; https://doi.org/10.1117/12.968230